HEMING YAO

PhD Candidate, Recipient of 2018 Graduate Student Award in Research

Bioinformatics


 

 

OGPS: What is your academic and research background?

Yao: I graduated from the University of Science and Technology of China, where I studied biological science and computer science. During my undergraduate study, I became interested in computational biology and medicine. Currently, I’m a PhD candidate in Bioinformatics. My research interests lie in applying and developing novel computational methods to solve biological and medical problems. When I was an undergraduate student, I worked on the application of advancement machine learning methods to study biomedical problems and create useful tooks including database and web servers to provide a prediction for post-translational modification site. During visits at the Harvard Medical School and Boston Children’s Hospital, I was actively involved in the building of an interface between healthcare data and innovation. In the bioinformatics program, I have continued to be dedicated to developing computational methods to address complex problems in medicine. I’m working on the left venticle segmentation of Short-axis Magnetic Resonance Imaging (MRI), and the hematoma segmentation in head Computed Tomography (CT) scans. Also, as a collaboration with DENSO, I’m studying drowsiness detection based on visual cues.

 

OGPS: How did you decide what you wanted to do in your science career?

Yao: My interest in bioinformatics started in the third year of my undergraduate program. I took a course called algorithm in medicine, where I realized how computational methods such as artificial intelligence could effectively help to solve complex problems in medicine. I started my learning by joining a bioinformatics lab, where I designed an effective name entity recognition method to extract disease name in literature automatically. Later, I joined a biomedical informatics lab at Boston Children’s Hospital, where I was trained in precision medicine. My role was to design an efficient data management structure to standardize clinical genetics data and implement Substitutable Medical Applications & Reusable Technologies (SMART) on the designed data structure. This work bridges the gap between clinical data in hospitals and genetics data in labs. This experience makes me realize the charm of bringing new techniques into real life. After these research experiences, I was inspired to continue working in this field and started my PhD. At UM, I joined the Biomedical & Clinical Informatics Laboratory and have been working on computational medicine.

 

OGPS: What are your current research initiatives?

Yao: I always think it’s essential to bring new techniques into real life. Only when the fancy algorithms are transferred into products, do they start to benefit people and society. At the Biomedical & Clinical Informatics Laboratory, I’m working on medical image analysis using deep learning techniques and other image processing/computer vision methods. As my lab has many industry collaborators, we actively push forward the methods in health informatics to be used and solve problems in practice. So, my research initiatives are to develop and deepen understanding of computational methods, which can facilitate the physician’s diagnosis and patient management.

OGPS: What accomplishment are you most proud of?

YaoThe accomplishment I’m most proud of is the method I proposed in the DENSO drowsiness detection project to reduce the computation cost and storage when using deep learning techniques. That method uses an automatic filter pruning to find and remove redundant kernels and speed up the network. This method is used to solve the problem that implementing deep learning techniques in the real-time application are often restricted by computation cost, memory storage, and energy efficiency. This challenge was proposed by our collaborators from DENSO when we were discussing the potentiality of our designed drowsiness detection system being used in cars. The development of this method involved discussion with researchers from both our lab and DENSO, literature survey, and dedicated network structure and experiment design. Finally, the proposed algorithm can efficiently speed up the inference time, and the patent on this method has been filled.

 

OGPS: What is one piece of advice you would give new PhD students?

Yao: Keep thinking. In a research project, wherever we are – literature review, designing the plan, implementing the experiment, or writing a manuscript, I think we should always actively think about what we are doing. Is the plan reasonable? Does the experimental design make sense? What is the potential pitfall? What are other people in this field doing? Sometimes, especially when we are in the middle of experiments, we may be trapped by some mistakes. Continuing thinking, discussing with colleagues, and reading the latest papers are things that always help us get through it. On the contrary, in research, we may waste time if we are working on assigned tasks without the brain. Also, when thinking about a struggling problem, it would be good to divide it into small particles and solve them one by one.

OGPS: What do you like to do outside of the lab?

Yao: I like traveling and watching movies in my free time. Both give me the opportunity to meet and know about something different.

OGPS: What are your future goals?

Yao: My long-term goal is to develop effective computational methods for biomedical data and implement an automated recommendation system that will facilitate the clinical prevention, diagnosis and patient care. To be more specific, my research interests lie in the application and development of computer vision and deep learning techniques in medical image/video analysis to extract powerful features and then integrate the derived features in the process of decision-making.